Autonomous QA Loop

🛡️ Autonomous Closed-Loop QA Engine

The Agentway QA ecosystem transforms testing from a passive manual check into an autonomous, self-healing closed loop.

---

🔁 The Closed-Loop Self-Healing Architecture

                       [Live Built Application]
                                  │
                                  ▼
             ┌────────────────────────────────────────┐
             │  Autonomous QA Evaluator Battery       │
             │  • prd-traceability-matrix             │
             │  • interactive-element-audit           │
             │  • flow-based-functional-testing       │
             │  • accessibility-runtime-audit (WCAG)  │
             │  • responsive-qa-audit (Playwright)    │
             └────────────────────┬───────────────────┘
                                  │
                                  ▼
             ┌────────────────────────────────────────┐
             │     qa-feedback-loop-orchestrator      │
             │     (Aggregates Findings & Auto-Fixes) │
             └────────────────────┬───────────────────┘
                                  │
                  ┌───────────────┴───────────────┐
                  │                               │
        [Deterministic Issue]           [Ambiguous Edge Case]
                  │                               │
                  ▼                               ▼
       Execute Auto-Fix Patch           Escalate to Human with
       (Max 3 Iterations Retry)         Complete Diagnostic Context

---

🎯 Deterministic Auto-Fix Rules

The orchestrator classifies QA findings into deterministic fixes vs human escalations:

Category	Finding Type	Action
Markup & A11y	Missing alt tag, aria-label, or unlinked <label>	Auto-Fix immediately
Touch Ergonomics	Tap target $&lt; 44 imes 44 ext{pt}$ or missing safe-area inset	Auto-Fix with token padding
Responsive Overflow	Horizontal scrollbar on mobile ($w \le 393 ext{px}$)	Auto-Fix max-w-full / flex wrap
Broken Routes	href="#" or unhandled button click handler	Auto-Fix route-intent mapping
Chaos & Edge Cases	Extreme long text overflow / unhandled negative numbers	Auto-Fix truncate / fallback —
Business Logic	Missing workflow step or conflicting PRD requirements	Escalate to Human

---

♿ WCAG 2.2 AA Compliance Coverage

The accessibility-runtime-audit executes axe-core checks covering:

1. Target Size (Minimum): $\ge 24 imes 24 ext{px}$ touch targets (WCAG 2.2 SC 2.5.8).
2. Focus Appearance: High contrast visible keyboard focus indicators.
3. Dragging Alternatives: Single-pointer alternatives for drag-and-drop operations.
4. Contrast (Minimum): $\ge 4.5:1$ text and $\ge 3:1$ UI components.

Generator #7: agentway doctor (v1.6.0)

Command: agentway doctor [dir] Alias: agentway health [dir]

Computes a Token Adherence Index (TAI) score from heuristic project signals and generates a three-part health report:

• health-report.md — Human-readable report card with radar breakdown and prescription table
• health-report.json — Machine-readable data for CI gate integration
• prescription.md — Prioritized (P0/P1/P2) skill activation roadmap

---

Generator #8: agentway audit:drift (v1.6.0)

Command: agentway audit:drift [dir] Alias: agentway drift [dir]

Scans all CSS/SCSS files in the target directory using 17 regex patterns across 7 drift categories. Severity-sorts violations and generates:

• drift-report.md — Violation list with file:line references and fix hints
• drift-report.json — Machine-readable data with per-category counts and drift rate
• drift-fixes.md — Project-specific find-and-replace auto-fix recipes